← CHI 2026 map

CHI '26 · Honorable mention · full-paper review · confidence medium-high

SoleCoach: Sole Pressure and IMU-based MLLMs for Skill Coaching

Toshihiro Hirano , Hitoshi Yoshihara , Yichen Peng , Chen-Chieh Liao , Erwin Wu , Hideki Koike

DOI PDF Program page Page JSON

SoleCoach is a credible CHI systems paper: the core move is to replace camera/pose dependence with insole pressure plus IMU and then let an MLLM turn those signals into coaching language. The novelty is strongest as a system architecture plus a new evaluation metric, while the evidence is bounded to a specific skiing context and learner population.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

SoleCoach’s contribution is best read as a pragmatic re-architecture of coaching feedback generation rather than a broad conceptual breakthrough. The paper’s central departure from common sense is that useful coaching can be produced from only insole pressure and IMU signals, without external cameras or motion capture, even in an outdoor, large-movement sport such as skiing. That is a meaningful systems claim because it changes the sensing assumptions that typically underpin coaching interfaces. The novelty is also not just in swapping sensors: the paper positions a multimodal large language model as the interpreter of those signals and adds AutoBCE, a body-part-oriented evaluation metric intended to judge whether feedback targets the right correction areas. That combination makes the contribution feel like a coherent pipeline rather than a single isolated trick. At the same time, the evidence supports a bounded claim set. The validation described in the provided material is mixed-methods in spirit: dataset collection, model evaluation, and a user study. But the user study is explicitly limited to six novice-to-intermediate skiers, so the strongest inference is about feasibility and perceived usefulness in that population, not universal coaching quality or expert-level training support. The limitations are important and credible: because the system only sees posture-related insole pressure and IMU data, it cannot capture environmental or contextual cues such as pole usage or turn timing relative to course layout. That means the method is well matched to independent practice and sensor-constrained settings, but not to every coaching scenario. Overall, this looks like a solid honorable-mention-level CHI systems contribution: technically inventive, well scoped, and useful, with claims that should remain tied to the specific skiing and wearable-sensing context rather than generalized too aggressively.

What Changed

Canon before

Prior CHI coaching-feedback systems largely depended on externally captured pose or video signals, which are less practical for outdoor sports and large-scale movement. This paper shifts the sensing assumption to insole pressure plus IMU and uses an MLLM to turn those signals into coaching language.

Departure from common sense

It is non-obvious that a coaching system can generate actionable skill feedback from only insole foot-pressure and IMU signals, without external cameras or body-mounted motion capture, especially for a sport like skiing where movement is large-scale and outdoor.

Actual novelty

The paper’s main novelty is a SoleCoach pipeline that interprets insole pressure and IMU signals with a multimodal large language model to produce ski coaching feedback, together with AutoBCE, a body-part-oriented evaluation metric for judging whether the generated coaching addresses the intended correction targets.

Evidence

The paper presents an insole-sensor-only coaching system for skiing, describes an MLLM-based interpretation pipeline, introduces AutoBCE as a new evaluation metric, and validates the approach with dataset collection plus a user study. The evidence supports a concrete system contribution and a bounded evaluation on novice-to-intermediate skiers, but not broad claims about all sports or athlete levels.

Limits

Method limits

The method depends on insole pressure and IMU signals, so it cannot directly observe external context such as pole use, course layout, or some timing cues. The evaluation also appears centered on a specific skiing setting and a limited participant pool, which constrains methodological generalization.

Deployment limits

Deployment is most plausible for independent practice scenarios where camera-based capture is impractical. It is less suitable when coaching requires environmental context, full-body kinematics, or expert assessment of advanced technique beyond what foot pressure and IMU can infer.

Boundary conditions

The paper’s own framing and evaluation suggest the strongest fit is alpine skiing practice with novice-to-intermediate learners using insole sensors. Claims should be bounded to feedback generation from sensor-derived posture cues rather than comprehensive sport-state understanding.

Manual check: No manual check required from the provided evidence packet, but the exact experimental details and any broader comparative claims should be verified against the full paper if publication decisions depend on them.

Position in field

This sits at the intersection of sports coaching, wearable sensing, and LLM-based feedback generation. Its field contribution is less about a new theory of coaching and more about a practical sensing-to-language system that relaxes the usual dependence on external pose capture.

Abstract

In sports training, individualized skill assessment and feedback are essential for athletes to master complex movements and enhance performance. Existing approaches for generating coaching comments primarily rely on externally captured pose information, which limits their applicability in outdoor sports such as skiing that involve large-scale movement. To address this challenge, we propose a method for presenting athletes' postures and generating coaching feedback solely based on foot pressure and IMU data collected from insole sensors. In our approach, a large language model directly interprets foot pressure signals to provide actionable coaching, thereby supporting independent practice. Through model evaluation and user studies, we demonstrate that the proposed method generates expert-level feedback and outperforms pose-based approaches. Furthermore, the user study shows that the feedback helps athletes identify body parts requiring correction and enhances their motivation for training.